From: Peter F. Patel-Schneider (pfps@research.bell-labs.com)
Date: 01/12/01
[I expect that I will also send this out to rdf-logic (possibly modified), but I wanted to send it to the committee now, and I can't send it to rdf-logic until DAML+OIL is released. peter] OK, now that there is a DAML+OIL release, it is time to try to put a spanner in the works. (Not that it is time to sabotage the release, it is just that it is now time to see what problems remain, so that the next release can fix them.) I expect that this list is not complete! Problem 1: The lack of a theory of identity (or equality, or equivalence) RDF and RDFS don't appear to have any need to determine whether two ``things'' are identical. They have no notion of equality, nor do they have any notion of cardinality or even uniqueness or functionality. However, DAML+OIL does include notions having to do with cardinality and uniqueness, so it needs to have access to a theory of identity. There are several aspects to the theory of identity (or equality) that is needed in DAML+OIL. There is the basic idea of when two ``normal'' resources are identical. (By normal resources, I mean resources that do not play any special role in DAML+OIL or RDF or RDFS.) There has been some talk about using a unique names assumption in DAML+OIL. I don't see how this can work because two different names (URIs) can refer to the same resource. We thus need to have some explicit notion of identity or equality in DAML+OIL, even for normal resources. There is the notion of identity of containers. Are two bags with the same members identical? Are they equal? Is there any difference between identity and equality here? Are two lists with the same elements in the same order identical? Are they equal? There is the notion of identity of statements. Are two statements with identital subjects, predicates, and objects identical? There is the notion of identity (and equality) of classes. Are two restrictions with the same content identical? Are they equal? Are two (non-restriction) classes with the same (non-primitive) definition equal? Are two classes with equivalent definitions equal? Are two classes that have the same members equal? Problem 2: Underspecified containers Even aside from problems of identity and equality, there are other problems with RDF containers. Can a container contain itself (directly or indirectly)? If so, the notions of identity (and equality) for containers become even more complex. Can a container be partially specified? If so, what does it mean? If not, how can it be prevented? Problem 3: The theory of classes. There are several aspects of classes that are underspecified in RDF and RDFS. Can a class contain itself? It has to, as a type of rdfs:Class is rdfs:Class. This means that determining whether two classes have the same members is quite complex. Can an entity be treated as both a class and an instance? In some sense it has to, because rdfs:Class is an instance of itself. Is a container a class? If so, are its elements instances of the class? Problem 4: The intent of RDF and RDFS and its interaction with DAML+OIL There are several aspects of RDF and RDFS that one might think affect DAML+OIL. For example, if we create a subclass of rdfs:Class and create an instance of this new subclass there is some expectation that we have just created a DAML+OIL class. However if we want this to happen, we have to formally specify the meaning of these portions of RDF and RDFS. The intended meanings of RDF and RDFS that may affect DAML+OIL include - the interaction between rdfs:subClass and rdf:type - the universality of rdfs:Resource - the meaning of distributive referents - the relationship between a statement and its reification
This archive was generated by hypermail 2.1.4 : 04/02/02 EST